A lyophilised peptide salt is an amorphous solid assembled by a route that is in contact with organic solvent at every stage, and some of that solvent can be retained in the solid. The reversed-phase chromatogram you run for purity will not tell you how much, because acetonitrile is the mobile phase and the other small solvents are unretained relative to it. A gravimetric drying loss will not tell you which solvent, because it weighs everything volatile at once and names nothing. Static headspace gas chromatography answers both questions, but it answers them through a partition equilibrium that your sample preparation controls far more tightly than your column does.
Where the solvent comes from
Fmoc solid-phase synthesis runs in dimethylformamide. It is the solvent the green-chemistry literature has been trying to displace, and displacing it is not a swap: a candidate replacement has to be requalified against amino acid solubility, resin swelling, deprotection kinetics and coupling before it can be claimed for all steps of SPPS [1]. Cleavage from the resin and preparative purification introduce the next two. Standard operating procedures for reversed-phase purification of peptides use acetonitrile as the organic modifier and trifluoroacetic acid as the ion-pairing reagent, over sequential C18, C4 and wide-pore diphenyl columns under gradient elution [2]. Because TFA serves as cleavage agent in the first step and ion-pairing reagent in the second, the isolated peptide is obtained as a TFA salt [3] — which matters here not as a solvent, but as a matrix property that will govern how the solid behaves in a hot headspace vial.
Lyophilisation is the last operation with a chance to remove anything volatile, and it is also where a solvent can be introduced on purpose. Freeze-drying from an organic cosolvent is used to change cake structure and crystallinity, and it leaves its own residue to be measured [15, 16].
The class scheme as a limit framework
ICH Q3C sorts solvents by toxicological concern rather than by chromatographic difficulty, which is why the hardest solvents to measure and the most tightly limited are not the same set. Five are designated Class 1, "should be avoided": benzene, carbon tetrachloride, 1,2-dichloroethane, 1,1-dichloroethene and 1,1,1-trichloroethane [4]. Treat the scheme as a framework that has been revised before rather than a physical constant. The Class 1 limits are unchanged from those originally proposed in 1997, and a detailed review of the toxicological information now available indicates a case for changing them for all five except benzene; the same review challenges the concept of expressing such limits as concentrations [4].
What the scheme sets for you at the bench is a target sensitivity, and generic methods are built to hit it across the Class 2 and Class 3 lists. One widely cited generic static headspace method covers 44 ICH Class 2 and 3 solvents in a 30-minute two-stage oven programme from 35 to 240 °C on a 30 m × 0.32 mm, 1.8 µm DB-624 column — and is explicitly not suitable for the 10 remaining Class 2 and 3 solvents, either because they are too polar (formic acid is the example given) or because they boil above 150 °C, such as anisole and cumene [6]. A generic method is a scope statement as much as a procedure. Read which solvents it excludes before you conclude that a solvent was absent.
The vial is a partition experiment
The sampler transfers gas. Any fraction of an analyte still dissolved in the liquid at the moment of sampling is invisible to the detector, so the quantity that determines your signal is the analyte's distribution between the diluent and the gas above it — and that distribution is a property of the diluent as much as of the analyte. This is the single most consequential fact in the method, and it is easy to demonstrate. Holding seventeen solvents used in drug substance synthesis and purification constant and varying only the matrix medium among DMSO, DMF, dimethylacetamide, benzyl alcohol, 1,3-dimethyl-2-imidazolidinone and water, peak shape for each analyte was unaffected while peak intensities for all solvents were strongly affected; the peak intensity of methanol in a benzyl alcohol matrix was more than four times that in a DMSO matrix [5].
Vial geometry and ionic strength move the same equilibrium. In a headspace SPME configuration, sample headspace volume had an important effect on sensitivity and precision: below one third of the vial volume sensitivity improved while precision worsened, and for 10 mL vials the optimum headspace volume was 3 mL; added Na2SO4 increased extraction yield, and slightly pressurising the vial improved sensitivity by a factor of two [10]. Those numbers belong to fibre extraction, not to loop transfer, so do not port them. Port the discipline: fill volume, salt content and vial lot are method parameters, and changing any of them re-scales your response.
Diluent choice, and the artifacts the diluent makes
Two constraints compete. The diluent must dissolve the sample, because an undissolved particle does not release occluded solvent into the liquid phase on the sampler's timetable; and it must not hold the analytes so tightly that nothing partitions out. DMSO is a common answer, chosen in one generic method for its high capacity to dissolve drug substance, its stability and its high boiling point [6]. A water–dimethylformamide mixture has been proposed instead as sample solvent to obtain good sensitivity and recovery, with selectivity checked against 36 solvents [7].
The diluent is also a reagent. In the matrix-medium study, artifacts were formed from the matrix media on heating in the headspace sampling apparatus, and also following ultrasonication of sample solutions used to increase dissolution of the sample into the medium; the artifacts included benzene and toluene, restricted as Class 1 and Class 2 solvents in the ICH guideline [5]. Sonicating a stubborn solid into a matrix medium is therefore a plausible way to manufacture an artifact peak that was never in the sample.
Salt form can catalyse this. During method development for formaldehyde, unusually high levels were seen with ethanol/DMSO (4:1, v/v) as diluent; replacing DMSO with dimethylformamide made formaldehyde undetectable, and a GC–MS experiment with DMSO-d6 yielding the expected diethoxymethane-d2 product confirmed that the drug substance, as its HCl salt, facilitated minor DMSO decomposition producing formaldehyde — an artifact peak rather than a true impurity. The same false positive appeared in four other compounds in HCl or HBr salt forms, which the authors suggest may indicate the phenomenon occurs widely during headspace GC of alkyl amines as HCl or HBr salts when DMSO-containing diluents are used [9]. A peptide TFA salt is not an alkyl amine hydrohalide, so this is not a prediction about your sample. It is the reason a diluent-only blank and a second diluent are cheap insurance whenever a small oxygenate appears.
The reverse interaction — matrix capturing analyte — is equally available. Volatile amines can chemically react with or interact with the sample matrix; in an acidic API, adding 1,8-diazabicyclo[5.4.0]undec-7-ene to a conventional high-boiling diluent drastically improved detectability and method accuracy for residual volatile amines, and also acted as a GC deactivation reagent that reduced interfacial adsorption of the analytes to GC components and improved precision [8]. That work was on basic analytes in an acidic matrix, not on neutral solvents, but it is the clearest published statement of the mechanism: if the liquid phase can hold the analyte chemically, headspace concentration is no longer a physical property you can calibrate against a clean standard.
Equilibration temperature and time
Temperature buys signal at a fixed rate over a usable range. Across the six matrix media tested, with a few exceptions, peak intensities approximately doubled for every 20 °C rise in equilibrium temperature between 80 and 140 °C, and there was no difference in this behaviour among the media [5]. Time buys completeness, not signal, once equilibrium is reached: in the generic method, equilibration temperature and time were assessed over 125–150 °C and 8–15 min, and residual solvents in 200 mg of drug substance were found to equilibrate efficiently at 140 °C for 10 min [6]. In the SPME method, the optimum equilibration time for all components and all systems was 30 min [10] — a reminder that the extraction geometry, not the analyte, sets the timescale.
The temperature that maximises signal is not automatically the temperature to run. The same heat that drives partition drives the diluent chemistry that generates artifacts [5, 9], so an equilibration study on a peptide should be read as two curves at once: analyte response rising, and blank response rising with it.
Calibrating against a matrix you cannot reproduce
External calibration assumes the analyte partitions the same way out of your standard as out of your sample. The matrix-medium data says that assumption is a choice you are making, not a fact you are relying on [5]. For a synthetic peptide the usual escape — prepare a blank matrix and spike it — is not available, because a peptide-free placebo with the same salt content, hygroscopicity and dissolution behaviour does not exist to buy.
Standard addition moves the calibration inside the sample. A static headspace method built to European Pharmacopoeia general procedure used a water–dimethylformamide mixture as sample solvent and the standard addition technique with internal standard quantitation for ethanol, tetrahydrofuran and toluene, validated to ICH Q2A and Q2B, with selectivity established across 36 solvents [7]. The cost is sample consumption and run time, since every determination needs its own addition series.
Platform methods are the alternative, and their validation boundary is the thing to read. A platform HS-GC procedure developed under the ICH Q14 enhanced approach, quantifying 18 residual solvents and resolving a critical pair with a confirmed method operable design region, focused method validation on the performance characteristics that do not require a sample matrix — specificity, range (linearity and quantitation limit) and reference solution stability — and provided a separate framework for adopting the platform to a specific product [11]. Inheriting that method gives you a validated separation. Accuracy in your matrix is still yours to demonstrate. The generic method makes the same point from the other side: recoveries of most of its solvents from four drug substances were greater than 80% [6] — four small molecules, none of them a peptide salt, and "most" rather than all.
FID, MS and the detector's blind spots
FID is the default because of sensitivity and dynamic range, but it responds exclusively to combustible organic species and produces no signal for common gases such as carbon dioxide, ammonia, and notably water [12]. It also returns retention time and area, not identity, which is a real exposure when a generic method meets a matrix it was not developed against. Mass spectrometric detection narrows that gap. In a static headspace method run with both detectors on botanical extracts spiked with a Class 3 solvent mix at 10 µg/g, MSD and FID recoveries were mainly 91–121% and 77–110% respectively in Coffeeberry extract, and 95–124% and 72–151% in pomegranate powder [14]. Those matrices are food botanicals rather than peptide salts and the absolute figures do not transfer; what transfers is the observation that the FID spread widened in the less characterised matrix, which is what a co-eluting matrix volatile looks like on a single-channel detector.
Thermal conductivity detection is the universal alternative, with a known response to all GC-appropriate compounds including water. A 7.5-minute headspace GC-TCD method quantitated water, over 25 common residual solvents and other volatile impurities in small-molecule pharmaceutical samples; with sample preparation controlled to minimise the impact of residual water from the diluent, the residual solvent results were comparable to GC-FID and the water results comparable to Karl Fischer [12].
Water, Karl Fischer, and why loss on drying under-reads
Karl Fischer measures the volatile the FID cannot see, and the two assays are complements rather than rivals. Headspace GC can be pushed onto water as well: a full evaporation technique weighs a small amount of solid directly into the vial and removes water thermally, which avoids searching for a diluent and avoids headspace saturation, with blank corrections applied to compensate for atmospheric moisture variation. With TCD the method gave R2 > 0.99 and RSD < 5% at each calibration level; eight samples were studied at different equilibration temperatures to find optimal conditions for each case, and the results were compared with Karl Fischer titration. Both methods showed restrictions, and proper case-by-case optimisation and validation turned out to be mandatory [13]. Hyphenating an FID in series with the TCD was useful for testing residual solvents simultaneously [13]. Nothing in that comparison retires Karl Fischer; it establishes that neither route is matrix-independent.
Loss on drying is a different kind of number: one gravimetric difference, which cannot say which volatile left and reports nothing at all for a volatile that does not leave under its conditions. That second failure mode is documented in freeze-dried solids. For tobramycin sulfate freeze-dried from tert-butanol/water, crystallisation of TBA before drying was critical to achieving adequately low residual TBA; rapid freezing left the TBA incompletely crystallised and gave relatively high residual solvent levels; annealing above the T'g of the system after an initial freezing step significantly reduced residual TBA; and secondary drying, even at increased temperature and for extended times, was not an effective method of reducing residual TBA levels [15]. In cephalothin sodium lyophilised from tert-butyl alcohol–water, gas chromatography found residual TBA at approximately 0.001% w/w in the crystalline lyophiles, and in general residual cosolvent levels were higher in lyophiles with lower crystallinity [16].
Neither system is a peptide and neither number transfers. What both establish, within their own systems, is that residual solvent in a freeze-dried solid tracked the physical state of that solid — crystallinity, thermal history — rather than how long it was dried. An amorphous lyophilised salt is the low-crystallinity case by construction, which is the structural reason a drying-loss specification is not a substitute for a headspace assay on such a solid: the headspace vial dissolves the solid at an elevated equilibration temperature in a diluent chosen to release the analyte, which is a fundamentally different release mechanism from evaporating from an intact cake in a drying oven.
What the number licenses
A static headspace result licenses one statement: this concentration of this named volatile, in this solid, under this diluent, this fill volume and this equilibration programme. It does not license a total-volatiles figure, because FID returns nothing for water [12]. It does not license a claim about anything non-volatile. And it does not license importing a recovery figure from someone else's validated matrix into yours [6, 11].
The trifluoroacetate counterion is outside its scope entirely, and is measured by different techniques: TFA- detection and quantification by FT-IR, 19F NMR and HPLC with evaporative light-scattering detection, all validated according to ICH guidelines, in a study that exchanged the counterion for chloride by freeze-drying from HCl and found 10 mM HCl the optimal condition with no impact on peptide purity observed at any HCl concentration tested [3]. Residual acetonitrile, residual TFA and water are three specifications answered by three different measurements. Reporting one and implying the others is the most common way this assay gets over-read.